Future Tense: Chapters 9–12

A lot can go awry with your inputs and with the hidden layers.

Theory of mind lets us intuit the mindset of others, and sometimes interpret and predict their behavior, without even being aware that we’re doing this.

It’s conceivable, though, that other errors like this might not be discovered. Obviously, diagnosing cancer is high stakes, which is why having clean data is paramount.

Many alignment problems get caught in the development phase: first, when programmers are making sure the goal is well defined; later, as they’re scrubbing errors and extraneous material from data; and finally, when ensuring the AI actually generated a solution that solves the problem it was meant to solve.

“Computers, if they are to be truly effective at decision making, will have to have emotions or emotion-like mechanisms,” Picard wrote in a groundbreaking book called Affective Computing.

This is where certain uses of affective computing algorithms are problematic.

As leaders, it is incumbent on all of us to make sure we are building a world in which every individual has an opportunity to thrive.

That’s not all it claims to do—it purports to calculate the likely gender of the person, the angle they’re looking at, and even the test-taker’s emotional state.

Honorlock acts as a proctor for students as they take tests on their laptops.

Teachers do a lot more for students than deliver curricula, of course; they work with students on social and emotional learning, guide them in academic collaborations, and show them in myriad ways how to work in a community.

Teachers do a lot more for students than deliver curricula, of course; they work with students on social and emotional learning, guide them in academic collaborations, and show them in myriad ways how to work in a community.

A doctor at Stanford University Hospital wondered if image analysis used to figure out the prognosis for lung cancer patients could also be tried on people with COVID-19.

First, machines analyzed CT-scan images, looking at lung lesions using something called automated image segmentation, which breaks an image into its related bits automatically with machine learning—kind of like identifying all the edge or corner pieces of a jigsaw puzzle.

But AI alone wasn’t as good as predictions paired with the results of lab tests and the clinical observations that human doctors make.

Meanwhile, researchers at Cornell are developing a prototype for a drone that they call a “flying guide dog.”